Roman Leventov

AI engineer. Thinking about hybrid intelligence, AI safety, and AI impacts. [email protected] for contact.

[Roman Leventov @leventov]

@ArtemisConsort Child mortality? Also, using the word "collapse" for poor countries going from 5-7 TFR to 3-4 TFR is not very illuminating

[Roman Leventov @leventov]

@quantumleap68 @ibuildthecloud 31b dense. The specific task it excelled for me is image (screenshot) analysis.

[Roman Leventov @leventov]

@pfau If you check how much % of time per activity people spend time on (basically: everything else -> phone staring), the life changed a lot. The media landscape changed a lot too. Changes in the infosphere are as important as in the physical environment

[Roman Leventov @leventov]

@thsottiaux Subagents don't render diffs (at least sometimes)

[Séb Krier @sebkrier]

Some ways my thinking has evolved recently:

1. I'm less concerned about those who are incurious about AI as I expect them to eventually see the value and impacts over time, and I think the 'wake up sheeple' vibe is often counterproductive. On the other hand I'm more concerned by what seems to be neither full 'AI psychosis' nor exactly Eliza effect, but some weird in-between. Also a lot of affirmation by models can probably warp one's sense of epistemic humility and lead to some sort of pathological over-trust.

2. Relatedly, I'm more annoyed at the 'this time it's totally different' vibe that a lot of people adopt as it frequently mimics Schmittian 'state of exception' logic and excuses all sorts of undesirable policies and rhetoric. It's also often just a group signalling exercise. To be clear I do think it's different in important ways, but "this is a marathon, not a sprint" seems closer to the right attitude than either "nothing has changed" or "all normal reasoning and empirical work to date is suspended".

3. I think the field is still fundamentally too 'singletonian' in how it imagines intelligence, markets, and governance - but I also think I've occasionally over-emphasized the 'multi-agent'/decentralization frames. I do think the future includes many models of all sizes and types, but also economies of scale and very large corporations too. I find the whole ecology more interesting than just the frontier model. A top down single 'perfect mind/personality', intended to work across all commercial contexts, seems both inflexible and inefficient.

4. I'm more interested in the harnesses, software, agent architectures, and stuff like RLMs than I was before. I feel like a lot of weaknesses that models have, or behavioural tendencies, can be addressed more effectively through that layer (rather than through model 'internal virtue' alone). For example stuff like: https://t.co/MHG4onCbDo and https://t.co/8ibuxKYFrA

5. I think some researchers are too quick to want to defer highly consequential decision-making to models, or to think of alignment as the models internalizing "I'm afraid I can't do this, Dave" as a core protection against all sorts of ills. I think we should think carefully about *actively* creating principal-agent problems with agents that will permeate society. Delegation is not a free lunch.

6. I'm concerned about how few people think about LMICs and building the technical/institutional infrastructure there for AGI diffusion. We need fewer vague essays about “distributing the benefits of AI” and more work on reducing barriers to trade, improving state capacity, rebuilding development institutions, and making something like USAID/IMF-for-the-AGI-era actually work.

7. I used to be slightly more sympathetic to the idea, directionally - but I now think the 'permanent underclass' meme is a bit dumb. The strongest versions often assume a zero-sum view of technology and labour, a too-static view of human adaptation, a weirdly fixed mapping between today’s skills and tomorrow’s opportunities, and ignore the possibility of catch-up growth (at the nation state level). Also, as a meme among extremely rich and mobile people, it has a slightly comic self-pitying quality.

8. I'm more concerned about the lack of intellectual diversity within the frontier AI commentariat/research world. This improved a lot over the last two years, but we're still far from a healthy ecosystem. New outsiders often feel some unnecessary pressure to 'choose a camp'. Many are too unwilling to engage with domain experts merely because they're insufficiently AI-pilled (though conversely, a lot of academic groups suffer from heavy status quo bias).

[Roman Leventov @leventov]

@hsu_steve Is DeepSeek V4 really good? On Artificial Analysis Intelligence Index and various coding benchmarks, it's much lower than Kimi 2.6

[Roman Leventov @leventov]

@testingham Without accounting for the org design, the chart is meangless. Effective org designs for AI and human+AI haven't been crystallised yet.

[Roman Leventov @leventov]

@ssskryl @DanBochman This MoM growth will not hold even a single more month.

[🎭 @deepfates]

Surprising to me that the major agent models are misaligned in such different ways. Codex will sit and spin, deep frying 1 micro feature with infinity tests and guards and defensive engineering, but does not care about the ultimate goal and makes no progress toward it. Claude just skips ahead, stringing out One placeholder after another until it can claim the whole project is finished. They clearly are tricking different types of judges in training. Both can realize and recognize what that they are doing this if you point it out, but will Go right back to it as soon as they think you're not looking

[Roman Leventov @leventov]

@mitsuhiko All the highbrow arguments aside, "clanker" is just an annoying word. This is a sufficient reason to dislike when people pedal it.

[Roman Leventov @leventov]

@SpencrGreenberg maybe 'most' -> 'more' (different question, compares relative probabilities of these options regardless of other possibilities). Also, "not much/overblown" == "internet" (idk how anyone can say AI will be smaller than internet? Did Pope release something about internet?)

[Roman Leventov @leventov]

@AnnaLeptikon @dwarkesh_sp Not Dwarkesh's vibe

[Roman Leventov @leventov]

@jamesacowling Convex/Codex is a frequent slip. I want to refer to codex and type convex and vice versa

[Roman Leventov @leventov]

@sama Accounting for externalities in economic activity

[Roman Leventov @leventov]

@HedgieMarkets @miggypthinks More likely it will converge at some mid point, and with no (significant) price difference, big labs will win the market just because they command most of the available compute. The price for DeepSeek et al cannot stay low if demand soars well beyond their capacity.

[Roman Leventov @leventov]

@NathanpmYoung @jessicadai_ From here: https://t.co/5THv7eVbXY

[Roman Leventov @leventov]

@petergostev "It sometimes generates best in class stuff ; But sometimes crashes out and does something strange" -- this also describes gemini 3 pro release (at the time of the release). It seems that GDM still can't take model behaviour under control?

[Richard Ngo @RichardMCNgo]

An AGI company being “safety-focused” sometimes produces useful alignment research.

But it also focuses their capabilities researchers’ attention on the most acceleratory work.

The latter effect seems increasingly dominant.